Building the actual skill in a fresh chat window instead of the one you used to test it is advice that goes against almost everything else I've read about this. Inheriting every wrong decision from the test ground makes obvious sense in hindsight, but I would have packaged it in-place without a second thought.
Yeah and that is the standard advice! So you wouldn’t be wrong. But when I used the same chat to package a skill I saw that a lot of the detours were also being noted down. As opposed to if I just ask it for a brief and do it in a new chat, the skill is actually cleaner.
I built an internal agent last year that I was convinced was "insanely good" because it nailed every test case I threw at it. Then I handed it to someone else on the team and it fell apart in about ten minutes, turns out I'd unconsciously trained my prompts around my own workflow quirks. Curious if you've hit that same wall: does "insanely good" hold up when someone other than the builder is driving it?
Yes! I have been open-sourcing my skills and agents for some time now. And I have been getting great reviews. The Reddit one that I open-sourced recently is helping people build their karma, promote their products, and find early users. But the method I use to build for myself vs to build for others is totally. What I described here is how I build it for myself. But when I do it for others, I make sure that there is a setup phase.
I recently started following you and this was a really helpful read.
I do have a question for you. I recently started using claude and for the last 2 months I've been creating projects ( instead of skills) for repeated tasks because claude projects can build context with the chat’s memory inside the project. And most of my job is to create content, scripts and pitch decks etc.
I can search for how both “projects” and “skills” are different but I want more of a user perspective on when to use what and efficiently.
Maybe you can publish one of the posts on this, if so that would be really helpful.
Okay so like you said, a project holds context. But a skill holds a process.
A project is where your files and background sit, so Claude starts every chat already knowing what you’re working on. A skill is the instruction set for how a task gets done, and Claude pulls it in whenever that task shows up (which Claude understands after a trigger word of sorts).
You can use Skills in a Project too. You can add them by going to your Settings -> under Customize -> Skills. And they apply everywhere. So you don’t have to pick one.
And for your work I would make skills for the repeated things, like how you write a script, craft a hook, add CTA, build a pitch deck, maybe for some research, etc. And I would keep projects for the context that changes, like a specific client.
But I barely use projects. My files and instructions live in Obsidian on my laptop, so skills plus Obsidian gets me better results than skills plus a project in my opinion.
And thanks for the topic suggestion! It is a great post idea. I will definitely write about this one!
Care should be taken for conscious use of Claude / its prompts, skills or agentic features. If something can be done outside ai - like normal Google search, pre research etc should be done as responsible citizen. Every action of ours will impact on environment- energy, water etc.
Building the actual skill in a fresh chat window instead of the one you used to test it is advice that goes against almost everything else I've read about this. Inheriting every wrong decision from the test ground makes obvious sense in hindsight, but I would have packaged it in-place without a second thought.
Yeah and that is the standard advice! So you wouldn’t be wrong. But when I used the same chat to package a skill I saw that a lot of the detours were also being noted down. As opposed to if I just ask it for a brief and do it in a new chat, the skill is actually cleaner.
I built an internal agent last year that I was convinced was "insanely good" because it nailed every test case I threw at it. Then I handed it to someone else on the team and it fell apart in about ten minutes, turns out I'd unconsciously trained my prompts around my own workflow quirks. Curious if you've hit that same wall: does "insanely good" hold up when someone other than the builder is driving it?
Yes! I have been open-sourcing my skills and agents for some time now. And I have been getting great reviews. The Reddit one that I open-sourced recently is helping people build their karma, promote their products, and find early users. But the method I use to build for myself vs to build for others is totally. What I described here is how I build it for myself. But when I do it for others, I make sure that there is a setup phase.
I recently started following you and this was a really helpful read.
I do have a question for you. I recently started using claude and for the last 2 months I've been creating projects ( instead of skills) for repeated tasks because claude projects can build context with the chat’s memory inside the project. And most of my job is to create content, scripts and pitch decks etc.
I can search for how both “projects” and “skills” are different but I want more of a user perspective on when to use what and efficiently.
Maybe you can publish one of the posts on this, if so that would be really helpful.
Hey! Thank you so much!!
Okay so like you said, a project holds context. But a skill holds a process.
A project is where your files and background sit, so Claude starts every chat already knowing what you’re working on. A skill is the instruction set for how a task gets done, and Claude pulls it in whenever that task shows up (which Claude understands after a trigger word of sorts).
You can use Skills in a Project too. You can add them by going to your Settings -> under Customize -> Skills. And they apply everywhere. So you don’t have to pick one.
And for your work I would make skills for the repeated things, like how you write a script, craft a hook, add CTA, build a pitch deck, maybe for some research, etc. And I would keep projects for the context that changes, like a specific client.
But I barely use projects. My files and instructions live in Obsidian on my laptop, so skills plus Obsidian gets me better results than skills plus a project in my opinion.
And thanks for the topic suggestion! It is a great post idea. I will definitely write about this one!
Thank you, I'll try this and also keep you updated, hope I can give you more content ideas😅.
Yes definitely! Haha thank you!
Care should be taken for conscious use of Claude / its prompts, skills or agentic features. If something can be done outside ai - like normal Google search, pre research etc should be done as responsible citizen. Every action of ours will impact on environment- energy, water etc.